Improving hydrological model identifiability by driving calibration with stochastic inputs

For a long time, the classical problem of identifying the optimal modeling structure and/or parameters followed the calibration-validation norm, originating from the iconic split-sample scheme by Vit Klemeš. A common feature of such approaches is their dependence on the length and representativeness of the available data. This introduces several questions since the inferred parameters are selected according to a specific subset (or subsets) of historical data, while the rest of data is used for validation. In this vein, we propose a conceptually simple approach driven by the well-known stochastic simulation paradigm, which builds upon the idea of calibrating models using alternative, yet probabilistically consistent, synthetic data. Decoupling this way, the available data now become the basis to generate stochastic inputs, as well as for model validation and parameter uncertainty assessment. This allows for embedding the stochasticity of real-world drivers (rainfall, evapotranspiration) and responses (runoff) and thus their hydrological uncertainty. Furthermore, it results to stable and robust models , as calibration is performed using long enough time series that reproduce important properties that are associated with the changing climate (e.g., long-term persistence), which are generally hidden in the short historical samples. Identifying this way, the derived parameters are optimal not only for the historical data set, but for any alternative plausible realization of the modeled processes.